2 * Copyright (c) 2010 The WebM project authors. All Rights Reserved.
4 * Use of this source code is governed by a BSD-style license
5 * that can be found in the LICENSE file in the root of the source
6 * tree. An additional intellectual property rights grant can be found
7 * in the file PATENTS. All contributing project authors may
8 * be found in the AUTHORS file in the root of the source tree.
12 * This is an example demonstrating multi-resolution encoding in VP8.
13 * High-resolution input video is down-sampled to lower-resolutions. The
14 * encoder then encodes the video and outputs multiple bitstreams with
15 * different resolutions.
22 #define VPX_CODEC_DISABLE_COMPAT 1
23 #include "vpx/vpx_encoder.h"
24 #include "vpx/vp8cx.h"
25 #include "vpx_ports/mem_ops.h"
26 #define interface (vpx_codec_vp8_cx())
27 #define fourcc 0x30385056
29 #define IVF_FILE_HDR_SZ (32)
30 #define IVF_FRAME_HDR_SZ (12)
33 * The input video frame is downsampled several times to generate a multi-level
34 * hierarchical structure. NUM_ENCODERS is defined as the number of encoding
35 * levels required. For example, if the size of input video is 1280x720,
36 * NUM_ENCODERS is 3, and down-sampling factor is 2, the encoder outputs 3
37 * bitstreams with resolution of 1280x720(level 0), 640x360(level 1), and
38 * 320x180(level 2) respectively.
40 #define NUM_ENCODERS 3
42 /* This example uses the scaler function in libyuv. */
43 #include "third_party/libyuv/include/libyuv/basic_types.h"
44 #include "third_party/libyuv/include/libyuv/scale.h"
45 #include "third_party/libyuv/include/libyuv/cpu_id.h"
47 static double vp8_mse2psnr(double Samples, double Peak, double Mse)
51 if ((double)Mse > 0.0)
52 psnr = 10.0 * log10(Peak * Peak * Samples / Mse);
54 psnr = 60; // Limit to prevent / 0
62 static void die(const char *fmt, ...) {
67 if(fmt[strlen(fmt)-1] != '\n')
72 static void die_codec(vpx_codec_ctx_t *ctx, const char *s) {
73 const char *detail = vpx_codec_error_detail(ctx);
75 printf("%s: %s\n", s, vpx_codec_error(ctx));
77 printf(" %s\n",detail);
81 int (*read_frame_p)(FILE *f, vpx_image_t *img);
83 static int read_frame(FILE *f, vpx_image_t *img) {
84 size_t nbytes, to_read;
87 to_read = img->w*img->h*3/2;
88 nbytes = fread(img->planes[0], 1, to_read, f);
89 if(nbytes != to_read) {
92 printf("Warning: Read partial frame. Check your width & height!\n");
97 static int read_frame_by_row(FILE *f, vpx_image_t *img) {
98 size_t nbytes, to_read;
102 for (plane = 0; plane < 3; plane++)
105 int w = (plane ? (1 + img->d_w) / 2 : img->d_w);
106 int h = (plane ? (1 + img->d_h) / 2 : img->d_h);
109 /* Determine the correct plane based on the image format. The for-loop
110 * always counts in Y,U,V order, but this may not match the order of
116 ptr = img->planes[img->fmt==VPX_IMG_FMT_YV12? VPX_PLANE_V : VPX_PLANE_U];
119 ptr = img->planes[img->fmt==VPX_IMG_FMT_YV12?VPX_PLANE_U : VPX_PLANE_V];
122 ptr = img->planes[plane];
125 for (r = 0; r < h; r++)
129 nbytes = fread(ptr, 1, to_read, f);
130 if(nbytes != to_read) {
133 printf("Warning: Read partial frame. Check your width & height!\n");
137 ptr += img->stride[plane];
146 static void write_ivf_file_header(FILE *outfile,
147 const vpx_codec_enc_cfg_t *cfg,
151 if(cfg->g_pass != VPX_RC_ONE_PASS && cfg->g_pass != VPX_RC_LAST_PASS)
157 mem_put_le16(header+4, 0); /* version */
158 mem_put_le16(header+6, 32); /* headersize */
159 mem_put_le32(header+8, fourcc); /* headersize */
160 mem_put_le16(header+12, cfg->g_w); /* width */
161 mem_put_le16(header+14, cfg->g_h); /* height */
162 mem_put_le32(header+16, cfg->g_timebase.den); /* rate */
163 mem_put_le32(header+20, cfg->g_timebase.num); /* scale */
164 mem_put_le32(header+24, frame_cnt); /* length */
165 mem_put_le32(header+28, 0); /* unused */
167 if(fwrite(header, 1, 32, outfile));
170 static void write_ivf_frame_header(FILE *outfile,
171 const vpx_codec_cx_pkt_t *pkt)
176 if(pkt->kind != VPX_CODEC_CX_FRAME_PKT)
179 pts = pkt->data.frame.pts;
180 mem_put_le32(header, pkt->data.frame.sz);
181 mem_put_le32(header+4, pts&0xFFFFFFFF);
182 mem_put_le32(header+8, pts >> 32);
184 if(fwrite(header, 1, 12, outfile));
187 int main(int argc, char **argv)
189 FILE *infile, *outfile[NUM_ENCODERS];
190 vpx_codec_ctx_t codec[NUM_ENCODERS];
191 vpx_codec_enc_cfg_t cfg[NUM_ENCODERS];
192 vpx_codec_pts_t frame_cnt = 0;
193 vpx_image_t raw[NUM_ENCODERS];
194 vpx_codec_err_t res[NUM_ENCODERS];
203 /*Currently, only realtime mode is supported in multi-resolution encoding.*/
204 int arg_deadline = VPX_DL_REALTIME;
206 /* Set show_psnr to 1/0 to show/not show PSNR. Choose show_psnr=0 if you
207 don't need to know PSNR, which will skip PSNR calculation and save
210 uint64_t psnr_sse_total[NUM_ENCODERS] = {0};
211 uint64_t psnr_samples_total[NUM_ENCODERS] = {0};
212 double psnr_totals[NUM_ENCODERS][4] = {{0,0}};
213 int psnr_count[NUM_ENCODERS] = {0};
215 /* Set the required target bitrates for each resolution level.
216 * If target bitrate for highest-resolution level is set to 0,
217 * (i.e. target_bitrate[0]=0), we skip encoding at that level.
219 unsigned int target_bitrate[NUM_ENCODERS]={1400, 500, 100};
220 /* Enter the frame rate of the input video */
222 /* Set down-sampling factor for each resolution level.
223 dsf[0] controls down sampling from level 0 to level 1;
224 dsf[1] controls down sampling from level 1 to level 2;
225 dsf[2] is not used. */
226 vpx_rational_t dsf[NUM_ENCODERS] = {{2, 1}, {2, 1}, {1, 1}};
228 if(argc!= (5+NUM_ENCODERS))
229 die("Usage: %s <width> <height> <infile> <outfile(s)> <output psnr?>\n",
232 printf("Using %s\n",vpx_codec_iface_name(interface));
234 width = strtol(argv[1], NULL, 0);
235 height = strtol(argv[2], NULL, 0);
237 if(width < 16 || width%2 || height <16 || height%2)
238 die("Invalid resolution: %ldx%ld", width, height);
240 /* Open input video file for encoding */
241 if(!(infile = fopen(argv[3], "rb")))
242 die("Failed to open %s for reading", argv[3]);
244 /* Open output file for each encoder to output bitstreams */
245 for (i=0; i< NUM_ENCODERS; i++)
247 if(!target_bitrate[i])
253 if(!(outfile[i] = fopen(argv[i+4], "wb")))
254 die("Failed to open %s for writing", argv[i+4]);
257 show_psnr = strtol(argv[NUM_ENCODERS + 4], NULL, 0);
259 /* Populate default encoder configuration */
260 for (i=0; i< NUM_ENCODERS; i++)
262 res[i] = vpx_codec_enc_config_default(interface, &cfg[i], 0);
264 printf("Failed to get config: %s\n", vpx_codec_err_to_string(res[i]));
270 * Update the default configuration according to needs of the application.
272 /* Highest-resolution encoder settings */
275 cfg[0].g_threads = 1; /* number of threads used */
276 cfg[0].rc_dropframe_thresh = 0;
277 cfg[0].rc_end_usage = VPX_CBR;
278 cfg[0].rc_resize_allowed = 0;
279 cfg[0].rc_min_quantizer = 4;
280 cfg[0].rc_max_quantizer = 56;
281 cfg[0].rc_undershoot_pct = 98;
282 cfg[0].rc_overshoot_pct = 100;
283 cfg[0].rc_buf_initial_sz = 500;
284 cfg[0].rc_buf_optimal_sz = 600;
285 cfg[0].rc_buf_sz = 1000;
286 //cfg[0].rc_dropframe_thresh = 10;
287 cfg[0].g_error_resilient = 1; /* Enable error resilient mode */
288 cfg[0].g_lag_in_frames = 0;
290 /* Disable automatic keyframe placement */
291 //cfg[0].kf_mode = VPX_KF_DISABLED;
292 cfg[0].kf_min_dist = cfg[0].kf_max_dist = 1000;
294 cfg[0].rc_target_bitrate = target_bitrate[0]; /* Set target bitrate */
295 cfg[0].g_timebase.num = 1; /* Set fps */
296 cfg[0].g_timebase.den = framerate;
298 /* Other-resolution encoder settings */
299 for (i=1; i< NUM_ENCODERS; i++)
301 memcpy(&cfg[i], &cfg[0], sizeof(vpx_codec_enc_cfg_t));
303 cfg[i].g_threads = 1; /* number of threads used */
304 cfg[i].rc_target_bitrate = target_bitrate[i];
306 /* Note: Width & height of other-resolution encoders are calculated
307 * from the highest-resolution encoder's size and the corresponding
308 * down_sampling_factor.
311 unsigned int iw = cfg[i-1].g_w*dsf[i-1].den + dsf[i-1].num - 1;
312 unsigned int ih = cfg[i-1].g_h*dsf[i-1].den + dsf[i-1].num - 1;
313 cfg[i].g_w = iw/dsf[i-1].num;
314 cfg[i].g_h = ih/dsf[i-1].num;
317 /* Make width & height to be multiplier of 2. */
318 // Should support odd size ???
319 if((cfg[i].g_w)%2)cfg[i].g_w++;
320 if((cfg[i].g_h)%2)cfg[i].g_h++;
323 /* Allocate image for each encoder */
324 for (i=0; i< NUM_ENCODERS; i++)
325 if(!vpx_img_alloc(&raw[i], VPX_IMG_FMT_I420, cfg[i].g_w, cfg[i].g_h, 32))
326 die("Failed to allocate image", cfg[i].g_w, cfg[i].g_h);
328 if (raw[0].stride[VPX_PLANE_Y] == raw[0].d_w)
329 read_frame_p = read_frame;
331 read_frame_p = read_frame_by_row;
333 for (i=0; i< NUM_ENCODERS; i++)
335 write_ivf_file_header(outfile[i], &cfg[i], 0);
337 /* Initialize multi-encoder */
338 if(vpx_codec_enc_init_multi(&codec[0], interface, &cfg[0], NUM_ENCODERS,
339 (show_psnr ? VPX_CODEC_USE_PSNR : 0), &dsf[0]))
340 die_codec(&codec[0], "Failed to initialize encoder");
342 /* The extra encoding configuration parameters can be set as follows. */
343 /* Set encoding speed */
344 for ( i=0; i<NUM_ENCODERS; i++)
347 if(vpx_codec_control(&codec[i], VP8E_SET_CPUUSED, speed))
348 die_codec(&codec[i], "Failed to set cpu_used");
350 /* Set static thresh for highest-resolution encoder. Set it to 1000 for
351 * better performance. */
353 unsigned int static_thresh = 1000;
354 if(vpx_codec_control(&codec[0], VP8E_SET_STATIC_THRESHOLD, static_thresh))
355 die_codec(&codec[0], "Failed to set static threshold");
357 /* Set static thresh = 0 for other encoders for better quality */
358 for ( i=1; i<NUM_ENCODERS; i++)
360 unsigned int static_thresh = 0;
361 if(vpx_codec_control(&codec[i], VP8E_SET_STATIC_THRESHOLD, static_thresh))
362 die_codec(&codec[i], "Failed to set static threshold");
368 while(frame_avail || got_data)
370 vpx_codec_iter_t iter[NUM_ENCODERS]={NULL};
371 const vpx_codec_cx_pkt_t *pkt[NUM_ENCODERS];
374 frame_avail = read_frame_p(infile, &raw[0]);
378 for ( i=1; i<NUM_ENCODERS; i++)
380 /*Scale the image down a number of times by downsampling factor*/
381 /* FilterMode 1 or 2 give better psnr than FilterMode 0. */
382 I420Scale(raw[i-1].planes[VPX_PLANE_Y], raw[i-1].stride[VPX_PLANE_Y],
383 raw[i-1].planes[VPX_PLANE_U], raw[i-1].stride[VPX_PLANE_U],
384 raw[i-1].planes[VPX_PLANE_V], raw[i-1].stride[VPX_PLANE_V],
385 raw[i-1].d_w, raw[i-1].d_h,
386 raw[i].planes[VPX_PLANE_Y], raw[i].stride[VPX_PLANE_Y],
387 raw[i].planes[VPX_PLANE_U], raw[i].stride[VPX_PLANE_U],
388 raw[i].planes[VPX_PLANE_V], raw[i].stride[VPX_PLANE_V],
389 raw[i].d_w, raw[i].d_h, 1);
393 /* Encode each frame at multi-levels */
394 if(vpx_codec_encode(&codec[0], frame_avail? &raw[0] : NULL,
395 frame_cnt, 1, flags, arg_deadline))
396 die_codec(&codec[0], "Failed to encode frame");
398 for (i=NUM_ENCODERS-1; i>=0 ; i--)
402 while( (pkt[i] = vpx_codec_get_cx_data(&codec[i], &iter[i])) )
405 switch(pkt[i]->kind) {
406 case VPX_CODEC_CX_FRAME_PKT:
407 write_ivf_frame_header(outfile[i], pkt[i]);
408 if(fwrite(pkt[i]->data.frame.buf, 1, pkt[i]->data.frame.sz,
411 case VPX_CODEC_PSNR_PKT:
416 psnr_sse_total[i] += pkt[i]->data.psnr.sse[0];
417 psnr_samples_total[i] += pkt[i]->data.psnr.samples[0];
418 for (j = 0; j < 4; j++)
420 //fprintf(stderr, "%.3lf ", pkt[i]->data.psnr.psnr[j]);
421 psnr_totals[i][j] += pkt[i]->data.psnr.psnr[j];
430 printf(pkt[i]->kind == VPX_CODEC_CX_FRAME_PKT
431 && (pkt[i]->data.frame.flags & VPX_FRAME_IS_KEY)? "K":".");
441 printf("Processed %ld frames.\n",(long int)frame_cnt-1);
442 for (i=0; i< NUM_ENCODERS; i++)
444 /* Calculate PSNR and print it out */
445 if ( (show_psnr) && (psnr_count[i]>0) )
448 double ovpsnr = vp8_mse2psnr(psnr_samples_total[i], 255.0,
451 fprintf(stderr, "\n ENC%d PSNR (Overall/Avg/Y/U/V)", i);
453 fprintf(stderr, " %.3lf", ovpsnr);
454 for (j = 0; j < 4; j++)
456 fprintf(stderr, " %.3lf", psnr_totals[i][j]/psnr_count[i]);
460 if(vpx_codec_destroy(&codec[i]))
461 die_codec(&codec[i], "Failed to destroy codec");
463 vpx_img_free(&raw[i]);
468 /* Try to rewrite the file header with the actual frame count */
469 if(!fseek(outfile[i], 0, SEEK_SET))
470 write_ivf_file_header(outfile[i], &cfg[i], frame_cnt-1);